1. Field of the Invention
The present invention relates to checkpoint security, and more particularly to an improved security and identification method and system for improving image quality in a facial recognition system for use in screening and/or identifying individuals passing through secure entry or checkpoints such as airport passenger terminals, government offices, office buildings, military facilities, laboratories, and other secure locations. It will be understood that the system and method disclosed herein can be used in improving image quality in other camera systems as well.
2. Description of the Prior Art
Recent national and international events have underscored the need for effective identification and screening of individuals entering or passing through secure checkpoints. Manual identification of persons at a checkpoint is a slow and cumbersome process, and because it is not precise, manual identification may allow unwanted or undesirable individuals to avoid detection through disguises. As a result, a number of closed-circuit television (CCTV) real-time facial recognition systems have been developed. These systems employ a computer software system that captures video images from a CCTV camera and compares the subject's facial features to those of the many subjects (e.g., terrorists or known felons) already stored in a computer database. In order for a system to function properly, it is important that the system receive a clear video image of the subject's face without heavy shadows or overexposed areas that may hide critical facial features and thereby make it difficult for the system to accurately recognize faces. It is therefore important that each subject passing through a secure checkpoint be briefly stopped and face or be directed to face in a particular direction for an optimal facial read. Because the nature of these systems is such that they may be installed in areas where lighting levels can change significantly (i.e. sunlight vs. clouds, day vs. night lighting, etc.), it is desirable to configure the camera to be less sensitive to external ambient lighting changes. To overcome the limitations of the prior art problems set forth above, such as the problem of assuring that the subject faces in an optimum direction and the problems attendant to variable ambient lighting conditions, the present invention discloses both apparatus and methods to accomplish these purposes.
Existing facial recognition or identification methods include: U.S. Pat. No. 3,805,238, which describes a primitive method for identifying individuals using selected characteristic body curves; U.S. Pat. No. 4,858,000, which describes an image recognition system designed for use in recognizing predetermined individuals out of a viewing audience; U.S. Pat. No. 5,163,094, which describes a method for identifying individuals by the analysis of elemental shapes derived from biosensor data, generating a thermal image of the individual's face that is converted to a digital representation; U.S. Pat. No. 5,905,807, which describes an apparatus for processing a facial image and creating an output of extracted data of feature points that may be used for comparison or other purposes; U.S. Pat. No. 6,111,517, which describes a continuous video monitoring system for regulating access to a restricted environment such as a computer system; and U.S. Pat. No. 6,292,575, which describes a real-time facial recognition and verification system for comparison to a stored image. Each of these patents describes either a method or apparatus for performing facial recognition or identification. However, all of these patents suffer from the drawback that unless the facial image is properly positioned and lighted for a sufficient length of time, the recognition or identification may be incomplete, inaccurate or may not take place at all.
U.S. Pat. No. 6,119,096 describes system using cameras deployed at a turnstile that scan the iris of a user for recognition. The system has sensors which detect the height of the subject, may instruct the subject to wait until scanning is completed before giving an instruction to proceed. U.S. Pat. No. 5,991,429 describes a facial recognition system for overt and covert use in which the camera angle may be adjusted to obtain a better image. However, neither of these patents discloses interactively instructing the subject for proper positioning, nor do they disclose any way to compensate for inadequate lighting.
The invention of U.S. Pat. No. 4,975,969 describes a method and apparatus for identifying individuals and encoding data onto security cards, and a related security system that uses the cards. In order to initialize a security card, the user inserts a blank card into a slot and the user's facial features are recorded in this “picture position” to establish a proper profile. The card reader at a security checkpoint is similarly placed so that the user automatically assumes the picture position at the checkpoint by inserting the card in the reader. If the cameras at the checkpoint do not get a good facial read, the user may be prompted to either reinsert the card or to push a button, causing the user to assume the picture position. It is apparent that this system requires and relies upon properly initialized security cards, and is not suitable for identification of subjects from the general population, such as at an airport check-in. The interactive instructions are only useful in such a pre-identification card-based system, and there is no disclosure of any way to compensate for inadequate lighting.
All of the above-described facial recognition systems suffer from common drawbacks that may prevent a complete or accurate identification/recognition of an individual subject passing through the checkpoint. A quickly moving individual may not remain in the field of view of the camera for a time period of sufficient length to enable proper facial scanning. As a result, an individual who is traveling quickly past the camera may not be recognized. Similarly, an individual may also escape recognition by the above-described system if the individual places a hand, arm, or other object in front of his or her face while passing by the camera. In addition, if there is inadequate ambient lighting in the vicinity of the checkpoint, even a high-resolution low light-sensitive camera may not be able to obtain a sufficiently clear image to perform complete or accurate identification and/or recognition.
Testing by the National Institute of Standards and Technology throughout the years had repeatedly concluded that the most difficult task for facial recognition systems is to match face images under uncontrolled lighting. In the 2002 Face Recognition Vendor Test the performance of the best system had dropped from 95% Verification Rate to 54% for the outdoor (uncontrolled lighting) test condition. Similarly, in the 2005 Face Recognition Grand Challenge the best system's performance decreased from 99% Verification Rate to below 80% for indoor images under uncontrolled lighting. These and other tests clearly showed that ambient lighting has a very significant negative effect on the performance of a face recognition system. Accordingly, there has been a long felt need in the industry for a recognition system that provides improved results under uncontrolled lighting conditions.